Leveraging Deep Reinforcement Learning for Cyber-Attack Paths Prediction: Formulation, Generalization, and Evaluation
Date:
In this talk, I presented my paper titled “Leveraging Deep Reinforcement Learning for Cyber-Attack Paths Prediction: Formulation, Generalization, and Evaluation.” We propose a novel approach that utilizes Deep Reinforcement Learning (DRL) techniques to approximate attacker decision-making. By adopting the attacker’s perspective and tactics, our method identifies potential attack paths, enabling proactive security analysis and the formulation of defense strategies.
Among the key contributions, we introduce a re-formulation of the problem where the DRL agent operates with a local view, selecting the source and target nodes for attacks at each timestep. Our training methodology further incorporates a diverse set of network topologies, varying in size and vulnerability, to enhance the model’s generalization capabilities.